GPT-4 Turbo

Faster, cheaper successor to GPT-4 with a larger context window

Paid Language
Visit Product Page →
128000 context tokens
Undisclosed parameters
Proprietary license
Nov 2023 released

GPT-4 Turbo is OpenAI’s November 2023 update to GPT-4, announced at the company’s first developer conference. It kept GPT-4’s underlying capability level but expanded the context window to 128,000 tokens, added a knowledge cutoff of April 2023, and cut pricing to $10 per million input tokens and $30 per million output tokens, well below original GPT-4 rates. It scores 86.5% on MMLU and 87.1% on HumanEval, and it added vision input, letting developers send images alongside text in the same request.

The larger context window came with a practical caveat: independent testing found that attention quality degraded noticeably past roughly 32,000 tokens, so the full 128K window worked better in theory than in dense, long-document use. Even so, GPT-4 Turbo became the default GPT-4-class model for API developers through much of 2024, available directly through OpenAI, through Azure OpenAI, and in ChatGPT, until GPT-4o replaced it as the recommended option later that year.